Graph isomorphism for Protein Active Site Detection

Elisa Cilia and Mauro Brunato

Information and Communication Technology,
University of Trento, via Sommarive, 14,
38100, Povo (Trento),
cilia|brunato@dit.unitn.it

EURO XXII, Prague, Czech Republic, July 2007

Abstract:
Thousands of protein structures have been resolved over recent years;
the problem of identifying the function of a protein can be modeled as
a classification task and solved in a supervised learning framework by
modeling a protein structure as a labeled graph. The similarity between
graph representations can be evaluated by formulating the problem as an
approximate graph or subgraph isomorphism keeping the representation
consistent with the biological structure. The efficiency of the technique,
based on reactive search mechanisms, is evaluated on the basis of the
classifier prediction accuracy.